rcutree.h 22 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510511512513514515516517518519520521522523524525526527528529530531532533534535536537538539540541542543544545546547548549550551552553554555556557558559560
  1. /*
  2. * Read-Copy Update mechanism for mutual exclusion (tree-based version)
  3. * Internal non-public definitions.
  4. *
  5. * This program is free software; you can redistribute it and/or modify
  6. * it under the terms of the GNU General Public License as published by
  7. * the Free Software Foundation; either version 2 of the License, or
  8. * (at your option) any later version.
  9. *
  10. * This program is distributed in the hope that it will be useful,
  11. * but WITHOUT ANY WARRANTY; without even the implied warranty of
  12. * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
  13. * GNU General Public License for more details.
  14. *
  15. * You should have received a copy of the GNU General Public License
  16. * along with this program; if not, write to the Free Software
  17. * Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA.
  18. *
  19. * Copyright IBM Corporation, 2008
  20. *
  21. * Author: Ingo Molnar <mingo@elte.hu>
  22. * Paul E. McKenney <paulmck@linux.vnet.ibm.com>
  23. */
  24. #include <linux/cache.h>
  25. #include <linux/spinlock.h>
  26. #include <linux/threads.h>
  27. #include <linux/cpumask.h>
  28. #include <linux/seqlock.h>
  29. /*
  30. * Define shape of hierarchy based on NR_CPUS, CONFIG_RCU_FANOUT, and
  31. * CONFIG_RCU_FANOUT_LEAF.
  32. * In theory, it should be possible to add more levels straightforwardly.
  33. * In practice, this did work well going from three levels to four.
  34. * Of course, your mileage may vary.
  35. */
  36. #define MAX_RCU_LVLS 4
  37. #define RCU_FANOUT_1 (CONFIG_RCU_FANOUT_LEAF)
  38. #define RCU_FANOUT_2 (RCU_FANOUT_1 * CONFIG_RCU_FANOUT)
  39. #define RCU_FANOUT_3 (RCU_FANOUT_2 * CONFIG_RCU_FANOUT)
  40. #define RCU_FANOUT_4 (RCU_FANOUT_3 * CONFIG_RCU_FANOUT)
  41. #if NR_CPUS <= RCU_FANOUT_1
  42. # define RCU_NUM_LVLS 1
  43. # define NUM_RCU_LVL_0 1
  44. # define NUM_RCU_LVL_1 (NR_CPUS)
  45. # define NUM_RCU_LVL_2 0
  46. # define NUM_RCU_LVL_3 0
  47. # define NUM_RCU_LVL_4 0
  48. #elif NR_CPUS <= RCU_FANOUT_2
  49. # define RCU_NUM_LVLS 2
  50. # define NUM_RCU_LVL_0 1
  51. # define NUM_RCU_LVL_1 DIV_ROUND_UP(NR_CPUS, RCU_FANOUT_1)
  52. # define NUM_RCU_LVL_2 (NR_CPUS)
  53. # define NUM_RCU_LVL_3 0
  54. # define NUM_RCU_LVL_4 0
  55. #elif NR_CPUS <= RCU_FANOUT_3
  56. # define RCU_NUM_LVLS 3
  57. # define NUM_RCU_LVL_0 1
  58. # define NUM_RCU_LVL_1 DIV_ROUND_UP(NR_CPUS, RCU_FANOUT_2)
  59. # define NUM_RCU_LVL_2 DIV_ROUND_UP(NR_CPUS, RCU_FANOUT_1)
  60. # define NUM_RCU_LVL_3 (NR_CPUS)
  61. # define NUM_RCU_LVL_4 0
  62. #elif NR_CPUS <= RCU_FANOUT_4
  63. # define RCU_NUM_LVLS 4
  64. # define NUM_RCU_LVL_0 1
  65. # define NUM_RCU_LVL_1 DIV_ROUND_UP(NR_CPUS, RCU_FANOUT_3)
  66. # define NUM_RCU_LVL_2 DIV_ROUND_UP(NR_CPUS, RCU_FANOUT_2)
  67. # define NUM_RCU_LVL_3 DIV_ROUND_UP(NR_CPUS, RCU_FANOUT_1)
  68. # define NUM_RCU_LVL_4 (NR_CPUS)
  69. #else
  70. # error "CONFIG_RCU_FANOUT insufficient for NR_CPUS"
  71. #endif /* #if (NR_CPUS) <= RCU_FANOUT_1 */
  72. #define RCU_SUM (NUM_RCU_LVL_0 + NUM_RCU_LVL_1 + NUM_RCU_LVL_2 + NUM_RCU_LVL_3 + NUM_RCU_LVL_4)
  73. #define NUM_RCU_NODES (RCU_SUM - NR_CPUS)
  74. extern int rcu_num_lvls;
  75. extern int rcu_num_nodes;
  76. /*
  77. * Dynticks per-CPU state.
  78. */
  79. struct rcu_dynticks {
  80. long long dynticks_nesting; /* Track irq/process nesting level. */
  81. /* Process level is worth LLONG_MAX/2. */
  82. int dynticks_nmi_nesting; /* Track NMI nesting level. */
  83. atomic_t dynticks; /* Even value for idle, else odd. */
  84. #ifdef CONFIG_RCU_FAST_NO_HZ
  85. int dyntick_drain; /* Prepare-for-idle state variable. */
  86. unsigned long dyntick_holdoff;
  87. /* No retries for the jiffy of failure. */
  88. struct timer_list idle_gp_timer;
  89. /* Wake up CPU sleeping with callbacks. */
  90. unsigned long idle_gp_timer_expires;
  91. /* When to wake up CPU (for repost). */
  92. bool idle_first_pass; /* First pass of attempt to go idle? */
  93. unsigned long nonlazy_posted;
  94. /* # times non-lazy CBs posted to CPU. */
  95. unsigned long nonlazy_posted_snap;
  96. /* idle-period nonlazy_posted snapshot. */
  97. int tick_nohz_enabled_snap; /* Previously seen value from sysfs. */
  98. #endif /* #ifdef CONFIG_RCU_FAST_NO_HZ */
  99. };
  100. /* RCU's kthread states for tracing. */
  101. #define RCU_KTHREAD_STOPPED 0
  102. #define RCU_KTHREAD_RUNNING 1
  103. #define RCU_KTHREAD_WAITING 2
  104. #define RCU_KTHREAD_OFFCPU 3
  105. #define RCU_KTHREAD_YIELDING 4
  106. #define RCU_KTHREAD_MAX 4
  107. /*
  108. * Definition for node within the RCU grace-period-detection hierarchy.
  109. */
  110. struct rcu_node {
  111. raw_spinlock_t lock; /* Root rcu_node's lock protects some */
  112. /* rcu_state fields as well as following. */
  113. unsigned long gpnum; /* Current grace period for this node. */
  114. /* This will either be equal to or one */
  115. /* behind the root rcu_node's gpnum. */
  116. unsigned long completed; /* Last GP completed for this node. */
  117. /* This will either be equal to or one */
  118. /* behind the root rcu_node's gpnum. */
  119. unsigned long qsmask; /* CPUs or groups that need to switch in */
  120. /* order for current grace period to proceed.*/
  121. /* In leaf rcu_node, each bit corresponds to */
  122. /* an rcu_data structure, otherwise, each */
  123. /* bit corresponds to a child rcu_node */
  124. /* structure. */
  125. unsigned long expmask; /* Groups that have ->blkd_tasks */
  126. /* elements that need to drain to allow the */
  127. /* current expedited grace period to */
  128. /* complete (only for TREE_PREEMPT_RCU). */
  129. atomic_t wakemask; /* CPUs whose kthread needs to be awakened. */
  130. /* Since this has meaning only for leaf */
  131. /* rcu_node structures, 32 bits suffices. */
  132. unsigned long qsmaskinit;
  133. /* Per-GP initial value for qsmask & expmask. */
  134. unsigned long grpmask; /* Mask to apply to parent qsmask. */
  135. /* Only one bit will be set in this mask. */
  136. int grplo; /* lowest-numbered CPU or group here. */
  137. int grphi; /* highest-numbered CPU or group here. */
  138. u8 grpnum; /* CPU/group number for next level up. */
  139. u8 level; /* root is at level 0. */
  140. struct rcu_node *parent;
  141. struct list_head blkd_tasks;
  142. /* Tasks blocked in RCU read-side critical */
  143. /* section. Tasks are placed at the head */
  144. /* of this list and age towards the tail. */
  145. struct list_head *gp_tasks;
  146. /* Pointer to the first task blocking the */
  147. /* current grace period, or NULL if there */
  148. /* is no such task. */
  149. struct list_head *exp_tasks;
  150. /* Pointer to the first task blocking the */
  151. /* current expedited grace period, or NULL */
  152. /* if there is no such task. If there */
  153. /* is no current expedited grace period, */
  154. /* then there can cannot be any such task. */
  155. #ifdef CONFIG_RCU_BOOST
  156. struct list_head *boost_tasks;
  157. /* Pointer to first task that needs to be */
  158. /* priority boosted, or NULL if no priority */
  159. /* boosting is needed for this rcu_node */
  160. /* structure. If there are no tasks */
  161. /* queued on this rcu_node structure that */
  162. /* are blocking the current grace period, */
  163. /* there can be no such task. */
  164. unsigned long boost_time;
  165. /* When to start boosting (jiffies). */
  166. struct task_struct *boost_kthread_task;
  167. /* kthread that takes care of priority */
  168. /* boosting for this rcu_node structure. */
  169. unsigned int boost_kthread_status;
  170. /* State of boost_kthread_task for tracing. */
  171. unsigned long n_tasks_boosted;
  172. /* Total number of tasks boosted. */
  173. unsigned long n_exp_boosts;
  174. /* Number of tasks boosted for expedited GP. */
  175. unsigned long n_normal_boosts;
  176. /* Number of tasks boosted for normal GP. */
  177. unsigned long n_balk_blkd_tasks;
  178. /* Refused to boost: no blocked tasks. */
  179. unsigned long n_balk_exp_gp_tasks;
  180. /* Refused to boost: nothing blocking GP. */
  181. unsigned long n_balk_boost_tasks;
  182. /* Refused to boost: already boosting. */
  183. unsigned long n_balk_notblocked;
  184. /* Refused to boost: RCU RS CS still running. */
  185. unsigned long n_balk_notyet;
  186. /* Refused to boost: not yet time. */
  187. unsigned long n_balk_nos;
  188. /* Refused to boost: not sure why, though. */
  189. /* This can happen due to race conditions. */
  190. #endif /* #ifdef CONFIG_RCU_BOOST */
  191. raw_spinlock_t fqslock ____cacheline_internodealigned_in_smp;
  192. } ____cacheline_internodealigned_in_smp;
  193. /*
  194. * Do a full breadth-first scan of the rcu_node structures for the
  195. * specified rcu_state structure.
  196. */
  197. #define rcu_for_each_node_breadth_first(rsp, rnp) \
  198. for ((rnp) = &(rsp)->node[0]; \
  199. (rnp) < &(rsp)->node[rcu_num_nodes]; (rnp)++)
  200. /*
  201. * Do a breadth-first scan of the non-leaf rcu_node structures for the
  202. * specified rcu_state structure. Note that if there is a singleton
  203. * rcu_node tree with but one rcu_node structure, this loop is a no-op.
  204. */
  205. #define rcu_for_each_nonleaf_node_breadth_first(rsp, rnp) \
  206. for ((rnp) = &(rsp)->node[0]; \
  207. (rnp) < (rsp)->level[rcu_num_lvls - 1]; (rnp)++)
  208. /*
  209. * Scan the leaves of the rcu_node hierarchy for the specified rcu_state
  210. * structure. Note that if there is a singleton rcu_node tree with but
  211. * one rcu_node structure, this loop -will- visit the rcu_node structure.
  212. * It is still a leaf node, even if it is also the root node.
  213. */
  214. #define rcu_for_each_leaf_node(rsp, rnp) \
  215. for ((rnp) = (rsp)->level[rcu_num_lvls - 1]; \
  216. (rnp) < &(rsp)->node[rcu_num_nodes]; (rnp)++)
  217. /* Index values for nxttail array in struct rcu_data. */
  218. #define RCU_DONE_TAIL 0 /* Also RCU_WAIT head. */
  219. #define RCU_WAIT_TAIL 1 /* Also RCU_NEXT_READY head. */
  220. #define RCU_NEXT_READY_TAIL 2 /* Also RCU_NEXT head. */
  221. #define RCU_NEXT_TAIL 3
  222. #define RCU_NEXT_SIZE 4
  223. /* Per-CPU data for read-copy update. */
  224. struct rcu_data {
  225. /* 1) quiescent-state and grace-period handling : */
  226. unsigned long completed; /* Track rsp->completed gp number */
  227. /* in order to detect GP end. */
  228. unsigned long gpnum; /* Highest gp number that this CPU */
  229. /* is aware of having started. */
  230. bool passed_quiesce; /* User-mode/idle loop etc. */
  231. bool qs_pending; /* Core waits for quiesc state. */
  232. bool beenonline; /* CPU online at least once. */
  233. bool preemptible; /* Preemptible RCU? */
  234. struct rcu_node *mynode; /* This CPU's leaf of hierarchy */
  235. unsigned long grpmask; /* Mask to apply to leaf qsmask. */
  236. #ifdef CONFIG_RCU_CPU_STALL_INFO
  237. unsigned long ticks_this_gp; /* The number of scheduling-clock */
  238. /* ticks this CPU has handled */
  239. /* during and after the last grace */
  240. /* period it is aware of. */
  241. #endif /* #ifdef CONFIG_RCU_CPU_STALL_INFO */
  242. /* 2) batch handling */
  243. /*
  244. * If nxtlist is not NULL, it is partitioned as follows.
  245. * Any of the partitions might be empty, in which case the
  246. * pointer to that partition will be equal to the pointer for
  247. * the following partition. When the list is empty, all of
  248. * the nxttail elements point to the ->nxtlist pointer itself,
  249. * which in that case is NULL.
  250. *
  251. * [nxtlist, *nxttail[RCU_DONE_TAIL]):
  252. * Entries that batch # <= ->completed
  253. * The grace period for these entries has completed, and
  254. * the other grace-period-completed entries may be moved
  255. * here temporarily in rcu_process_callbacks().
  256. * [*nxttail[RCU_DONE_TAIL], *nxttail[RCU_WAIT_TAIL]):
  257. * Entries that batch # <= ->completed - 1: waiting for current GP
  258. * [*nxttail[RCU_WAIT_TAIL], *nxttail[RCU_NEXT_READY_TAIL]):
  259. * Entries known to have arrived before current GP ended
  260. * [*nxttail[RCU_NEXT_READY_TAIL], *nxttail[RCU_NEXT_TAIL]):
  261. * Entries that might have arrived after current GP ended
  262. * Note that the value of *nxttail[RCU_NEXT_TAIL] will
  263. * always be NULL, as this is the end of the list.
  264. */
  265. struct rcu_head *nxtlist;
  266. struct rcu_head **nxttail[RCU_NEXT_SIZE];
  267. unsigned long nxtcompleted[RCU_NEXT_SIZE];
  268. /* grace periods for sublists. */
  269. long qlen_lazy; /* # of lazy queued callbacks */
  270. long qlen; /* # of queued callbacks, incl lazy */
  271. long qlen_last_fqs_check;
  272. /* qlen at last check for QS forcing */
  273. unsigned long n_cbs_invoked; /* count of RCU cbs invoked. */
  274. unsigned long n_nocbs_invoked; /* count of no-CBs RCU cbs invoked. */
  275. unsigned long n_cbs_orphaned; /* RCU cbs orphaned by dying CPU */
  276. unsigned long n_cbs_adopted; /* RCU cbs adopted from dying CPU */
  277. unsigned long n_force_qs_snap;
  278. /* did other CPU force QS recently? */
  279. long blimit; /* Upper limit on a processed batch */
  280. /* 3) dynticks interface. */
  281. struct rcu_dynticks *dynticks; /* Shared per-CPU dynticks state. */
  282. int dynticks_snap; /* Per-GP tracking for dynticks. */
  283. /* 4) reasons this CPU needed to be kicked by force_quiescent_state */
  284. unsigned long dynticks_fqs; /* Kicked due to dynticks idle. */
  285. unsigned long offline_fqs; /* Kicked due to being offline. */
  286. /* 5) __rcu_pending() statistics. */
  287. unsigned long n_rcu_pending; /* rcu_pending() calls since boot. */
  288. unsigned long n_rp_qs_pending;
  289. unsigned long n_rp_report_qs;
  290. unsigned long n_rp_cb_ready;
  291. unsigned long n_rp_cpu_needs_gp;
  292. unsigned long n_rp_gp_completed;
  293. unsigned long n_rp_gp_started;
  294. unsigned long n_rp_need_nothing;
  295. /* 6) _rcu_barrier() and OOM callbacks. */
  296. struct rcu_head barrier_head;
  297. #ifdef CONFIG_RCU_FAST_NO_HZ
  298. struct rcu_head oom_head;
  299. #endif /* #ifdef CONFIG_RCU_FAST_NO_HZ */
  300. /* 7) Callback offloading. */
  301. #ifdef CONFIG_RCU_NOCB_CPU
  302. struct rcu_head *nocb_head; /* CBs waiting for kthread. */
  303. struct rcu_head **nocb_tail;
  304. atomic_long_t nocb_q_count; /* # CBs waiting for kthread */
  305. atomic_long_t nocb_q_count_lazy; /* (approximate). */
  306. int nocb_p_count; /* # CBs being invoked by kthread */
  307. int nocb_p_count_lazy; /* (approximate). */
  308. wait_queue_head_t nocb_wq; /* For nocb kthreads to sleep on. */
  309. struct task_struct *nocb_kthread;
  310. #endif /* #ifdef CONFIG_RCU_NOCB_CPU */
  311. int cpu;
  312. struct rcu_state *rsp;
  313. };
  314. /* Values for fqs_state field in struct rcu_state. */
  315. #define RCU_GP_IDLE 0 /* No grace period in progress. */
  316. #define RCU_GP_INIT 1 /* Grace period being initialized. */
  317. #define RCU_SAVE_DYNTICK 2 /* Need to scan dyntick state. */
  318. #define RCU_FORCE_QS 3 /* Need to force quiescent state. */
  319. #define RCU_SIGNAL_INIT RCU_SAVE_DYNTICK
  320. #define RCU_JIFFIES_TILL_FORCE_QS 3 /* for rsp->jiffies_force_qs */
  321. #define RCU_STALL_RAT_DELAY 2 /* Allow other CPUs time */
  322. /* to take at least one */
  323. /* scheduling clock irq */
  324. /* before ratting on them. */
  325. #define rcu_wait(cond) \
  326. do { \
  327. for (;;) { \
  328. set_current_state(TASK_INTERRUPTIBLE); \
  329. if (cond) \
  330. break; \
  331. schedule(); \
  332. } \
  333. __set_current_state(TASK_RUNNING); \
  334. } while (0)
  335. /*
  336. * RCU global state, including node hierarchy. This hierarchy is
  337. * represented in "heap" form in a dense array. The root (first level)
  338. * of the hierarchy is in ->node[0] (referenced by ->level[0]), the second
  339. * level in ->node[1] through ->node[m] (->node[1] referenced by ->level[1]),
  340. * and the third level in ->node[m+1] and following (->node[m+1] referenced
  341. * by ->level[2]). The number of levels is determined by the number of
  342. * CPUs and by CONFIG_RCU_FANOUT. Small systems will have a "hierarchy"
  343. * consisting of a single rcu_node.
  344. */
  345. struct rcu_state {
  346. struct rcu_node node[NUM_RCU_NODES]; /* Hierarchy. */
  347. struct rcu_node *level[RCU_NUM_LVLS]; /* Hierarchy levels. */
  348. u32 levelcnt[MAX_RCU_LVLS + 1]; /* # nodes in each level. */
  349. u8 levelspread[RCU_NUM_LVLS]; /* kids/node in each level. */
  350. struct rcu_data __percpu *rda; /* pointer of percu rcu_data. */
  351. void (*call)(struct rcu_head *head, /* call_rcu() flavor. */
  352. void (*func)(struct rcu_head *head));
  353. #ifdef CONFIG_RCU_NOCB_CPU
  354. void (*call_remote)(struct rcu_head *head,
  355. void (*func)(struct rcu_head *head));
  356. /* call_rcu() flavor, but for */
  357. /* placing on remote CPU. */
  358. #endif /* #ifdef CONFIG_RCU_NOCB_CPU */
  359. /* The following fields are guarded by the root rcu_node's lock. */
  360. u8 fqs_state ____cacheline_internodealigned_in_smp;
  361. /* Force QS state. */
  362. u8 boost; /* Subject to priority boost. */
  363. unsigned long gpnum; /* Current gp number. */
  364. unsigned long completed; /* # of last completed gp. */
  365. struct task_struct *gp_kthread; /* Task for grace periods. */
  366. wait_queue_head_t gp_wq; /* Where GP task waits. */
  367. int gp_flags; /* Commands for GP task. */
  368. /* End of fields guarded by root rcu_node's lock. */
  369. raw_spinlock_t orphan_lock ____cacheline_internodealigned_in_smp;
  370. /* Protect following fields. */
  371. struct rcu_head *orphan_nxtlist; /* Orphaned callbacks that */
  372. /* need a grace period. */
  373. struct rcu_head **orphan_nxttail; /* Tail of above. */
  374. struct rcu_head *orphan_donelist; /* Orphaned callbacks that */
  375. /* are ready to invoke. */
  376. struct rcu_head **orphan_donetail; /* Tail of above. */
  377. long qlen_lazy; /* Number of lazy callbacks. */
  378. long qlen; /* Total number of callbacks. */
  379. /* End of fields guarded by orphan_lock. */
  380. struct mutex onoff_mutex; /* Coordinate hotplug & GPs. */
  381. struct mutex barrier_mutex; /* Guards barrier fields. */
  382. atomic_t barrier_cpu_count; /* # CPUs waiting on. */
  383. struct completion barrier_completion; /* Wake at barrier end. */
  384. unsigned long n_barrier_done; /* ++ at start and end of */
  385. /* _rcu_barrier(). */
  386. /* End of fields guarded by barrier_mutex. */
  387. atomic_long_t expedited_start; /* Starting ticket. */
  388. atomic_long_t expedited_done; /* Done ticket. */
  389. atomic_long_t expedited_wrap; /* # near-wrap incidents. */
  390. atomic_long_t expedited_tryfail; /* # acquisition failures. */
  391. atomic_long_t expedited_workdone1; /* # done by others #1. */
  392. atomic_long_t expedited_workdone2; /* # done by others #2. */
  393. atomic_long_t expedited_normal; /* # fallbacks to normal. */
  394. atomic_long_t expedited_stoppedcpus; /* # successful stop_cpus. */
  395. atomic_long_t expedited_done_tries; /* # tries to update _done. */
  396. atomic_long_t expedited_done_lost; /* # times beaten to _done. */
  397. atomic_long_t expedited_done_exit; /* # times exited _done loop. */
  398. unsigned long jiffies_force_qs; /* Time at which to invoke */
  399. /* force_quiescent_state(). */
  400. unsigned long n_force_qs; /* Number of calls to */
  401. /* force_quiescent_state(). */
  402. unsigned long n_force_qs_lh; /* ~Number of calls leaving */
  403. /* due to lock unavailable. */
  404. unsigned long n_force_qs_ngp; /* Number of calls leaving */
  405. /* due to no GP active. */
  406. unsigned long gp_start; /* Time at which GP started, */
  407. /* but in jiffies. */
  408. unsigned long jiffies_stall; /* Time at which to check */
  409. /* for CPU stalls. */
  410. unsigned long gp_max; /* Maximum GP duration in */
  411. /* jiffies. */
  412. char *name; /* Name of structure. */
  413. struct list_head flavors; /* List of RCU flavors. */
  414. };
  415. /* Values for rcu_state structure's gp_flags field. */
  416. #define RCU_GP_FLAG_INIT 0x1 /* Need grace-period initialization. */
  417. #define RCU_GP_FLAG_FQS 0x2 /* Need grace-period quiescent-state forcing. */
  418. extern struct list_head rcu_struct_flavors;
  419. /* Sequence through rcu_state structures for each RCU flavor. */
  420. #define for_each_rcu_flavor(rsp) \
  421. list_for_each_entry((rsp), &rcu_struct_flavors, flavors)
  422. /* Return values for rcu_preempt_offline_tasks(). */
  423. #define RCU_OFL_TASKS_NORM_GP 0x1 /* Tasks blocking normal */
  424. /* GP were moved to root. */
  425. #define RCU_OFL_TASKS_EXP_GP 0x2 /* Tasks blocking expedited */
  426. /* GP were moved to root. */
  427. /*
  428. * RCU implementation internal declarations:
  429. */
  430. extern struct rcu_state rcu_sched_state;
  431. DECLARE_PER_CPU(struct rcu_data, rcu_sched_data);
  432. extern struct rcu_state rcu_bh_state;
  433. DECLARE_PER_CPU(struct rcu_data, rcu_bh_data);
  434. #ifdef CONFIG_TREE_PREEMPT_RCU
  435. extern struct rcu_state rcu_preempt_state;
  436. DECLARE_PER_CPU(struct rcu_data, rcu_preempt_data);
  437. #endif /* #ifdef CONFIG_TREE_PREEMPT_RCU */
  438. #ifdef CONFIG_RCU_BOOST
  439. DECLARE_PER_CPU(unsigned int, rcu_cpu_kthread_status);
  440. DECLARE_PER_CPU(int, rcu_cpu_kthread_cpu);
  441. DECLARE_PER_CPU(unsigned int, rcu_cpu_kthread_loops);
  442. DECLARE_PER_CPU(char, rcu_cpu_has_work);
  443. #endif /* #ifdef CONFIG_RCU_BOOST */
  444. #ifndef RCU_TREE_NONCORE
  445. /* Forward declarations for rcutree_plugin.h */
  446. static void rcu_bootup_announce(void);
  447. long rcu_batches_completed(void);
  448. static void rcu_preempt_note_context_switch(int cpu);
  449. static int rcu_preempt_blocked_readers_cgp(struct rcu_node *rnp);
  450. #ifdef CONFIG_HOTPLUG_CPU
  451. static void rcu_report_unblock_qs_rnp(struct rcu_node *rnp,
  452. unsigned long flags);
  453. #endif /* #ifdef CONFIG_HOTPLUG_CPU */
  454. static void rcu_print_detail_task_stall(struct rcu_state *rsp);
  455. static int rcu_print_task_stall(struct rcu_node *rnp);
  456. static void rcu_preempt_check_blocked_tasks(struct rcu_node *rnp);
  457. #ifdef CONFIG_HOTPLUG_CPU
  458. static int rcu_preempt_offline_tasks(struct rcu_state *rsp,
  459. struct rcu_node *rnp,
  460. struct rcu_data *rdp);
  461. #endif /* #ifdef CONFIG_HOTPLUG_CPU */
  462. static void rcu_preempt_check_callbacks(int cpu);
  463. void call_rcu(struct rcu_head *head, void (*func)(struct rcu_head *rcu));
  464. #if defined(CONFIG_HOTPLUG_CPU) || defined(CONFIG_TREE_PREEMPT_RCU)
  465. static void rcu_report_exp_rnp(struct rcu_state *rsp, struct rcu_node *rnp,
  466. bool wake);
  467. #endif /* #if defined(CONFIG_HOTPLUG_CPU) || defined(CONFIG_TREE_PREEMPT_RCU) */
  468. static void __init __rcu_init_preempt(void);
  469. static void rcu_initiate_boost(struct rcu_node *rnp, unsigned long flags);
  470. static void rcu_preempt_boost_start_gp(struct rcu_node *rnp);
  471. static void invoke_rcu_callbacks_kthread(void);
  472. static bool rcu_is_callbacks_kthread(void);
  473. #ifdef CONFIG_RCU_BOOST
  474. static void rcu_preempt_do_callbacks(void);
  475. static int __cpuinit rcu_spawn_one_boost_kthread(struct rcu_state *rsp,
  476. struct rcu_node *rnp);
  477. #endif /* #ifdef CONFIG_RCU_BOOST */
  478. static void __cpuinit rcu_prepare_kthreads(int cpu);
  479. static void rcu_prepare_for_idle_init(int cpu);
  480. static void rcu_cleanup_after_idle(int cpu);
  481. static void rcu_prepare_for_idle(int cpu);
  482. static void rcu_idle_count_callbacks_posted(void);
  483. static void print_cpu_stall_info_begin(void);
  484. static void print_cpu_stall_info(struct rcu_state *rsp, int cpu);
  485. static void print_cpu_stall_info_end(void);
  486. static void zero_cpu_stall_ticks(struct rcu_data *rdp);
  487. static void increment_cpu_stall_ticks(void);
  488. static bool is_nocb_cpu(int cpu);
  489. static bool __call_rcu_nocb(struct rcu_data *rdp, struct rcu_head *rhp,
  490. bool lazy);
  491. static bool rcu_nocb_adopt_orphan_cbs(struct rcu_state *rsp,
  492. struct rcu_data *rdp);
  493. static bool nocb_cpu_expendable(int cpu);
  494. static void rcu_boot_init_nocb_percpu_data(struct rcu_data *rdp);
  495. static void rcu_spawn_nocb_kthreads(struct rcu_state *rsp);
  496. static void init_nocb_callback_list(struct rcu_data *rdp);
  497. static void __init rcu_init_nocb(void);
  498. #endif /* #ifndef RCU_TREE_NONCORE */
  499. #ifdef CONFIG_RCU_TRACE
  500. #ifdef CONFIG_RCU_NOCB_CPU
  501. /* Sum up queue lengths for tracing. */
  502. static inline void rcu_nocb_q_lengths(struct rcu_data *rdp, long *ql, long *qll)
  503. {
  504. *ql = atomic_long_read(&rdp->nocb_q_count) + rdp->nocb_p_count;
  505. *qll = atomic_long_read(&rdp->nocb_q_count_lazy) + rdp->nocb_p_count_lazy;
  506. }
  507. #else /* #ifdef CONFIG_RCU_NOCB_CPU */
  508. static inline void rcu_nocb_q_lengths(struct rcu_data *rdp, long *ql, long *qll)
  509. {
  510. *ql = 0;
  511. *qll = 0;
  512. }
  513. #endif /* #else #ifdef CONFIG_RCU_NOCB_CPU */
  514. #endif /* #ifdef CONFIG_RCU_TRACE */